Thank you Matt, valid and important point. The architecture goes beyond replacing RAM with memristors - it introduces a fundamentally different physical substrate and* computational model*. Unlike traditional digital neural networks, which simulate activity symbolically on von Neumann machines, *EDI <https://bit.ly/45JPjsg>* is grounded in continuous spatial dynamics driven by real ionic and charge interactions akin to what occurs in biological neurons, see this paper https://www.nature.com/articles/npre.2010.5345.2
The distinction is not just in the hardware, it is in the *computational paradigm*: information is processed and stored in the same medium through dynamic field interactions, not separated across memory and processing units. This allows true material-based learning and self-organization, which digital LLMs do not display. Regarding simulations, initial models are currently in development. However, the goal is not to simulate EDI in its entirety, as many of its properties cannot be meaningfully captured in a digital environment. EDI fundamentally departs from conventional AI architectures. Muchlike you can't simulate flight in a way that generates real lift, some properties of EDI, e.g. emergent spatiotemporal behavior, only manifest in physical substrates - a new class of intelligence *from the physics of the system itself.* By using LLMs initially to pre-instantiate the early stages of intelligence within an EDI framework (see papers's paragraph) , you can train the system efficiently.If you replace the LLM's memory architecture with memristors, you begin to approach the architecture we envisioned in the manuscript Electrodynamic Intelligence (EDI). Once EDI develops its own internal adaptive dynamics, grounded in physical memory, coupling, and energy-efficient computation, the need for LLMs themselves diminishes. As biologically fragile organisms, we face significant limitations when it comes to colonizing Mars or other planetary environments. To thrive beyond Earth, we’ll need systems that can adapt, and build autonomously in extreme conditions. We need an *Optimus 5.0 *equipped with an EDI brain, vital for off-world infrastructure, habitat construction, and autonomous problem-solving - today feels almost alien intelligence. *Wouldn't that feel like having an LLM in the 1940s? * With targeted investment from both public and private sectors, a functional *EDI prototype* could realistically be developed within 2–3 years, maybe less given the current pace of innovation - --Dorian Aur On Sun, Aug 24, 2025 at 11:05 AM Matt Mahoney <[email protected]> wrote: > Do you have any experimental data or simulations using memristors that > prove your electrodynamic theory of intelligence? I realize that > neuromorphic computing with tungsten titanium oxide memristors has > demonstrated handwritten digit recognition. > https://www.nature.com/articles/s41528-024-00356-6 > > And that hafnium zirconium oxide memristors could also be used, as both > technologies are compatible with CMOS chip manufacturing. But it seems to > me we are maybe a decade away from large scale manufacturing as problems > with quality control are worked out. > https://www.nature.com/articles/s41467-025-61758-2 > > Maybe this will solve the power problem. WO3 memristors work by moving > oxygen vacancies, which I guess requires less energy than moving electrons > into a DRAM capacitor because atoms are heavier and slower. > > But what's missing from your claim is evidence that intelligence is > fundamentally different than current neural networks used in LLMs. My > understanding is we are just replacing RAM with memristors. Do you have any > simulations supporting your proposed architecture, which is not clear from > your paper? > > -- Matt Mahoney, [email protected] > > On Sun, Aug 24, 2025, 12:21 PM Dorian Aur <[email protected]> wrote: > >> >> For over 80 years, from the abstractions of McCulloch and Pitts to >> today’s large language models, we have been building simulated >> intelligence. Today's "AI" is a digital replica of brain-like processes, >> running on silicon and operating through mathematical operations instead of >> biological ones. >> >> Think of it like a flight simulator. A simulator can accurately recreate >> the cockpit experience and train pilots - it will never fly. *The >> simulator never actually leaves the ground* . Digital AI is similar: it >> emulates brain intelligence however lacks a true physical embodiment. >> >> This is where *Electrodynamic Intelligence (EDI) >> <https://doi.org/10.5281/zenodo.16929461>* offers a transformative path >> forward. >> >> Rather than modeling intelligence through symbolic or statistical >> computation, in *EDI cognition develops from real-time physical >> interactions, e.g. self-organizing dynamics of charges and fields within >> and across neurons in an artificial brain*. The electrodynamic >> processes are not metaphors; they are materially grounded forms of >> computation that operate through ionic flows, field interactions, and >> nonlinear dynamics. >> >> *EDI <https://bit.ly/45JPjsg>* is not a simulation of the brain, it is >> an *embodied approach to intelligence*, rooted in the same physical >> principles that underlie the biological brain. It opens a path toward >> systems that *act* through matter rather than *model* through >> abstraction. >> >> Just as real flight requires lift, drag, and thrust, not numbers on a >> screen, this* embodied intelligence requires the physics of the brain, >> not just the logic of the code*. Ben, I hope we’ll have the opportunity >> to feature a summary of this paper at the upcoming AGI conference >> >> Fifteen years ago, when we launched *Neuroelectrodynamics* as a >> theoretical framework, the technological landscape simply wasn’t ready to >> support its implementation. However now, *Colin*, the situation has >> changed dramatically, we're in a position to build. >> >> - Dorian Aur >> >> PS Matt, EDI doesn’t just solve these problems, it avoids them >> altogether by not sharing the same structural assumptions. It’s not >> about controlling artificial goals, it is about building intelligence that >> grows, adapts, and lives in the world as we do, not above it. >> >> >> [image: image.png] >> >> >> >> On Mon, Aug 11, 2025 at 9:46 AM Matt Mahoney <[email protected]> >> wrote: >> >>> Discussion of AI existential risk on LessWrong. To summarize: we don't >>> know how to solve the alignment problem. If we build AGI, it will probably >>> kill all humans because we dont know how to give it the right goals. >>> Therefore we should not build it, or at least build an "off" switch to >>> quickly shut it down. >>> >>> My thoughts: >>> >>> 1. The premise seems correct. We measure intelligence by prediction >>> accuracy. Wolpert's law says two agents cannot mutually predict each other. >>> If an agent is smarter than you, then you can't predict its actions, and >>> therefore cannot control it. >>> >>> 2. An LLM has no goals. It just predicts text. However, applications >>> that use it do have goals. You can tell an LLM to express any human goals >>> or feelings. So alignment seems solvable, at least for now. >>> >>> 3. Let's say we do solve the alignment problem. Then AGI will kill us by >>> giving us everything we want. AI agents will replace not just workers, but >>> friends and lovers too. We will become socially isolated and stop having >>> children. >>> >>> 4. The goal of all agents in a finite universe is a state of maximum >>> utilitiy, where any thought or perception is unpleasant because it would >>> result in a different state. Your goal is death. You just don't know it >>> because evolution programmed you to fear death. >>> >>> 5. An "off" switch will fail because AGI could kill us before we knew >>> anything was wrong. I don't even know why they proposed it. >>> >>> 6. We will build AGI anyway because human labor costs $50 trillion per >>> year, half of global GDP. >>> >>> >>> *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > <https://agi.topicbox.com/groups/agi/Ta9b77fda597cc07a-M93ed36a63dce68e320198528> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Ta9b77fda597cc07a-Mca135dad0ce59fb6532c2ed3 Delivery options: https://agi.topicbox.com/groups/agi/subscription
